WeeklyTelcon_20240130

Jump to bottom

Tommy Janjusic edited this page Jan 30, 2024 · 9 revisions

Open MPI Weekly Telecon ---

Dialup Info: (Do not post to public mailing list or public wiki)

Attendees (on Web-ex)

Tommy Janjusic (NVIDIA)
Christoph Niethammer (HLRS)
Joseph Schuchart (UTK)
Todd Kordenbrock
Luke Robison (AWS)
Jeff Squyres (CISCO)
Edgar Gabriel (AMD)
Thomas Huber (Cornelis)
Wenduo Wang (AWS)
David Bernholdt (ORNL)
Thomas Naughton (ORNL)

v4.1.x

v4.1.x Issues; v4.1.x Questions
- 12270: btl smcuda hang in v4.1.5
v4.1.x Open PRs
Next release end of Q1.

v5.0.x

New issues since v5.0.1
Open PRs since v5.0.1
Closed/Merged PRs since v5.0.1
v5.0.x Open PRs
- Non-critical changes will be held back until 5.0.2 release
- 12135, will merge after 5.0.2.
v5.0.x Issues; v5.0.x Questions
v5.0.2 delayed

Main branch

Additional Agenda Items

Jeff: Progress on testing using automation to close tickets where we need user to reply, but they don't reply
- Use two different bots:
  - https://github.com/marketplace/actions/close-stale-issues
  - https://github.com/marketplace/actions/close-issues-after-no-reply
- Requirements:
  - Only do this automation on github issues where we apply a specific label (e.g., "needs-reply")
  - If anyone replies who does not have write perms to the repo, automation remotes the "needs-reply" label
  - If X time goes by with no replies, bot adds a polite comment "We're still waiting for a reply; if Y more time goes by with no reply, we'll close this issue"
  - If Y more time goes by with no replies, bot adds a polite comment "We're assuming this issue has been abandoned, and will close it", and then actually closes the issue.
- Testing on Jeff's fork.
  - Issue with no label, so it won't go stale/close
  - Issue with label, so it should get stale and eventually close
  - Issue showing automation remove the label when user replied to the issue -- should go stale/auto-close because I re-applied the needs-reply label
Luke: Looking for ideas on how we can cache SMSC endpoints for use in hierarchy-aware collectives. Draft PR
Edgar: face-to-face engineering meeting in Spring in Austin?

TODO

Notes

Slow MPI_Group_difference · Issue #12286 · open-mpi/ompi (github.com) Short explanation no promises

V4.1 btl_smcuda hang - reproduce in main? Luke is looking into it next week Maybe Edgar can also reproduce, will check

[v4.1.x] fix atomic operations opal_atomic_compare_exchange_strong_ in arm64 by yuncliu · Pull Request #12005 · open-mpi/ompi (github.com) keep open until resolution on main

^ critical, but not marked as blocker yet

Newer versions of OpenMPI are unable to locate CUDA support. · Issue #12264 · open-mpi/ompi (github.com)

Config/opal_check_cuda check for with-cuda-libdir

smsc/xpmem: Refactor reg-cache to use tree's find() instead of iterate() by gkatev · Pull Request #11358 · open-mpi/ompi (github.com) Probably needs merged, and maybe solve: WIP: Add smsc endpoints to HAN by lrbison · Pull Request #12272 · open-mpi/ompi (github.com) Luke will take a look at smsc reg

Clone this wiki locally